n8n Pro MCP

Permanently delete a project by ID — cannot be undone. Move its workflows/credentials out first (transfer_workflow / transfer_credential).

Delete a tag by ID (removes it from every workflow; the workflows themselves are untouched).

Permanently remove a user from the instance by ID or email — cannot be undone; the user loses access immediately.

Permanently delete a variable by ID — cannot be undone. Workflows reading $vars.<key> will get undefined afterwards.

Permanently delete a workflow by ID. This cannot be undone — consider deactivate_workflow instead if you only want to stop it.

Generate n8n's built-in security audit: flags abandoned workflows, unused credentials, risky nodes (filesystem/instance access), and more. Optionally restrict categories or tune the abandoned-workflow threshold.

Get the JSON schema of a credential type — which fields it needs (e.g. 'slackApi' → accessToken). Use before create_credential.

Get one execution by ID. includeData=true returns the full run data (node inputs/outputs and error details) — useful to debug a failed run.

Aggregate execution health over a recent sample: per-workflow success/error/waiting counts, success rate and average duration. The queue-mode dashboard in one call — spot failing or slow workflows.

Get one user by ID or email address.

Get a workflow by ID, including full nodes and connections JSON. Use mode='summary' to get only metadata and the node list (cheaper for inspection).

List the tags attached to a workflow.

List executions with filters (status, workflow, project) and automatic pagination. Returns lightweight summaries with duration — use get_execution for full node-level data.

List projects on the instance (requires the Projects feature, available on licensed plans). Project IDs are used by transfer_workflow / transfer_credential and as filters.

Queue-mode view: list executions currently running or waiting in the queue. Tries the API status filter first and falls back to scanning recent executions for ones without a stop time.

List all tags on the instance with their IDs (IDs are needed by set_workflow_tags).

List users of the instance with their global roles (requires owner API key).

List instance variables (accessible in workflows via $vars.<key>). Requires a licensed n8n plan with the Variables feature.

List workflows with optional filters (active, tag names, project, name) and automatic cursor pagination. Returns id, name, active state, tags and timestamps — never the heavy node data.

Bulk-delete finished execution records by filter (status, workflow, older than N days) to reclaim database space. Runs as a dry run by default — set dryRun=false to actually delete. Deletion is permanent and cannot be undone. Running/waiting executions are never touched.

Pull changes from the connected source-control repository (requires the Source Control feature configured on the instance). force=true permanently discards local changes on the instance — cannot be undone.

Remove a user from a project (the user keeps their instance account).

Retry a failed or stopped execution by ID. Requires a recent n8n version exposing retry in the public API — if unavailable (404), re-trigger the workflow via run_webhook instead.

Trigger a workflow through its Webhook node. Calls {instance}/webhook/{path} (or /webhook-test/{path} with test=true, which requires the workflow open in 'Listen for test event' mode). In queue mode the call returns according to the webhook's response mode.

Search workflows by free text across name, node names and node types (client-side over the listing — finds e.g. every workflow using a given node type or touching a given service).

Replace the tags of a workflow. Accepts tag NAMES — resolves them to IDs automatically and, with createMissing=true, creates tags that don't exist yet. Pass an empty list to remove all tags.

Bulk activate or deactivate every workflow carrying a tag — an operational kill switch (e.g. deactivate everything tagged 'client-x'). Runs as a dry run by default; set dryRun=false to apply. Workflows already in the desired state are skipped.

Debug a failed execution without flooding the context: fetches the full run data and returns only the failing node, error message/description, a short stack and per-node errors. Use instead of get_execution includeData=true when you only need to know what broke.

Transfer a credential to another project (requires the Projects feature).

Transfer a workflow to another project (requires the Projects feature on the instance).

Rename a project by ID.

Rename a tag by ID (all workflows using it see the new name).

Update an existing variable's key/value by variable ID (see list_variables for IDs).

Update a workflow partially: fetches the current version, merges only the fields you pass (name, nodes, connections, settings), validates offline and PUTs the result. Pass full nodes/connections arrays when changing them — they replace the existing ones.

Validate a workflow JSON offline (no API call): structure, unique node names, connection integrity, orphan nodes, {{ }} expression syntax, webhook `.body` access, Code node return format and hardcoded secrets. Run before create_workflow/update_workflow.

Poll an execution until it reaches a terminal state (success/error/crashed/canceled) or the timeout expires. Returns early with waiting=true if the execution pauses on a Wait node. Closes the loop after run_webhook in queue mode, where execution happens asynchronously on workers. The timeout is best-effort: each poll request has its own HTTP timeout, so total wall time can slightly exceed timeoutSeconds.